Amanda Lewis

Hello, and thank you to Gene and the whole ETLS team for creating this space — a safe space where we can share our stories.

So tonight I'm going to share some confessions with you. Most of my career, I've been entrusted with the responsibility of promoting continuous improvement.

So my first confession is a bit embarrassing.

I do not like change.

I've been known to say, "Do as I say, not as I do." But as I was writing this talk, I realized I don't think that's actually true. I am actually more of a slow adopter. I love to read, I love to learn, I love to experiment with those learnings. But it kind of depends on what that change is. And if it's in my workflow, and it's not my idea, I kind of sit on that resistance phase of the change curve for quite a bit. So I think really — I'm a cautious adopter.

So we get to where I was earlier this year. My mind was filled with several topics, and everywhere I turned, generative AI was in the discussion. For over a year, I've been super excited about GenAI. I've been experimenting with it to help our users. But to put it in my workflow — where would I find the time to do this? What if it takes more work? And how long can I actually avoid this?

Well, it was actually the DORA community discussion that got me to a place where I was ready. Once I changed my mindset that I didn't have to do this, but I could experiment with it, my curiosity really took over. It kept honing in on the findings from last year's research around code reviews.

So I decided I'll experiment with GenAI in my developer workflow to see how it impacts documentation, testing, knowledge sharing, and code maintainability. I only have five minutes, so I'm going to share a few more confessions based on the outcomes of my experience.

How I got started was — I pulled in open source samples. I had no experience with this code, and in some cases I had no experience with the language. Hello, Java. Using Gemini Code Assist, I would have it explain the files, explain the project, and sometimes specific pieces of the code. I was really surprised at how quickly I could get up to speed on the project.

DORA started researching documentation in 2021, and year after year we see that the research found that quality internal documentation enables teams to improve their software delivery. So I understand how important documentation is. Quality internal documentation — it's a game changer. But I always seem to put it at the bottom of the to-do list, because it doesn't come naturally to me, and I can really stare at a blank piece of paper for a very long time.

So before I share the learnings from my experiments, if we could all just take a moment and remember all of that documentation I planned to write but I didn't. And you'll notice it ends in 2024. And that's because with Code Assist, I no longer have this blank sheet of paper. I ask it to help me write the documentation. I open it in a diff file, and then I actually have fun going through the suggestions. Sometimes it needs editing. Sometimes I find something in my code I didn't expect, and then I can fix it before I ask for a review.

I've also experimented with using it to have it write the documentation that could be used in an internal wiki. Now I find myself using this not just for code I write, but when I'm exploring other code as well.

So pair programming definitely invokes some strong opinions — you are either in favor or you're against it. Well, now I actually find myself wanting to pair program. I still have some anxiety, but I find it enjoyable, and it gives me a lot of energy interacting with Code Assist. I became more comfortable with asking questions — better conversational questions, not how I would go and search in Google or Stack Overflow. And I found that it feels easier to have meaningful conversations with my teammates about generated code. It leads to discussions that end up — we share more context about the code base and ideas for improvement.

So with all the new technologies, I was feeling overwhelmed. But now, after going through these experiments, I find myself — I'll go be out for a walk, I get an idea, I can't wait to get back to my computer and see if I can figure it out.

Yes, I definitely found positive impacts on those capabilities that drive improving your code review speed. But more importantly, I improved my well-being — and I didn't expect that, and I'm super excited about it. So with more context and less cognitive load, we will all have more time and energy for ideation and collaboration.

Do you have confessions, learnings, or surprises you want to share? If you're a change agent and a cautious adopter, come see me. Let's talk about it. Woo.